N Gram Overlap

You suspect your custom eval set leaked into a model's training data. Describe two practical tests you would run to detect contamination.

Explain benchmark contamination and its effect on reported LLM capability scores